Site-to-Site VPN Between Two Google Cloud Regions
What you build: HA VPN with BGP between two Google Cloud networks in two regions, failover when a tunnel dies, and a wrong-secret fix. Every step tested.

✍️ Translate this blog into Hindi, Tamil or Telugu
A site-to-site VPN joins two private networks over the internet so their machines talk on private addresses, encrypted. Companies use it to connect an office or a data centre to the cloud. In this lab you build Google Cloud's HA VPN between two separate networks in two regions (playing "head office" and "branch"), let Cloud Router learn the routes with BGP, prove traffic flows, then pull a tunnel out and give another one the wrong secret to see what high availability really means.
Who it is for: cloud and network engineers who want to see VPN and BGP work, not just read about them. Time: about 1 hour. Cost: HA VPN tunnels and two small VMs for about 20 minutes, covered by the free trial credit. Delete everything at the end: tunnels cost by the hour.
Every command below was run for real by the CareerByteCode Team, and every output is the real result we got. We ran the gcloud commands from a Cloud Shell-like Linux machine; Cloud Shell in your browser works the same.
Part A: VPN, HA VPN and BGP in plain English

Words you will use:
- VPC network: a private network in Google Cloud. Ours:
vpc-office(10.10.0.0/24, us-east1) andvpc-branch(10.20.0.0/24, us-central1). The ranges must not overlap. - HA VPN gateway: Google's VPN endpoint. It has two interfaces, each with its own public IP, so one can fail while the other carries traffic.
- Tunnel: one encrypted connection between two interfaces, set up with IKE (the handshake) and a shared secret both sides know. We build four: two from each side.
- Cloud Router + BGP: instead of typing routes by hand, each Cloud Router tells the other which ranges it can reach (BGP, the routing protocol of the internet). Each router has an ASN (a number that names it in BGP): 65001 and 65002.
- Link addresses: each BGP session talks over a tiny private range like 169.254.0.0/30, inside the tunnel.
- Real-life example: HA VPN is two guarded bridges between two cities. BGP is the traffic office on each side that tells drivers which streets lie across the river. If one bridge closes, traffic uses the other.
Before you start
A Google Cloud project with billing linked (the free trial works) and Cloud Shell open, with PROJECT_ID set to your project id and the Compute Engine API on.
Part B: the lab
Step 1: two separate networks
Why: the two networks play two sites. Each gets one subnet and a firewall rule that allows ping and SSH from the other site's range and from Google's IAP range (for SSH without public IPs).
# Two separate networks that play "head office" (us-east1) and "branch" (us-central1). Their ranges must not overlap
gcloud compute networks create vpc-office --subnet-mode=custom > /dev/null && gcloud compute networks subnets create office-sub --network=vpc-office --region=us-east1 --range=10.10.0.0/24 > /dev/null
gcloud compute networks create vpc-branch --subnet-mode=custom > /dev/null && gcloud compute networks subnets create branch-sub --network=vpc-branch --region=us-central1 --range=10.20.0.0/24 > /dev/null
for n in office branch; do gcloud compute firewall-rules create $n-allow-vpn --network=vpc-$n --allow=icmp,tcp:22 --source-ranges=10.10.0.0/24,10.20.0.0/24,35.235.240.0/20 > /dev/null; done
gcloud compute networks subnets list --filter="name~^(office|branch)-sub" --format="table(name,region.basename(),ipCidrRange)"Created [https://www.googleapis.com/compute/v1/projects/your-project-id/global/networks/vpc-office].
Instances on this network will not be reachable until firewall rules
are created. As an example, you can allow all internal traffic between
instances as well as SSH, RDP, and ICMP by running:
$ gcloud compute firewall-rules create <FIREWALL_NAME> --network vpc-office --allow tcp,udp,icmp --source-ranges <IP_RANGE>
$ gcloud compute firewall-rules create <FIREWALL_NAME> --network vpc-office --allow tcp:22,tcp:3389,icmp
Created [https://www.googleapis.com/compute/v1/projects/your-project-id/regions/us-east1/subnetworks/office-sub].
Created [https://www.googleapis.com/compute/v1/projects/your-project-id/global/networks/vpc-branch].
Instances on this network will not be reachable until firewall rules
are created. As an example, you can allow all internal traffic between
instances as well as SSH, RDP, and ICMP by running:
$ gcloud compute firewall-rules create <FIREWALL_NAME> --network vpc-branch --allow tcp,udp,icmp --source-ranges <IP_RANGE>
$ gcloud compute firewall-rules create <FIREWALL_NAME> --network vpc-branch --allow tcp:22,tcp:3389,icmp
Created [https://www.googleapis.com/compute/v1/projects/your-project-id/regions/us-central1/subnetworks/branch-sub].
Creating firewall...
..Created [https://www.googleapis.com/compute/v1/projects/your-project-id/global/firewalls/office-allow-vpn].
done.
Creating firewall...
..Created [https://www.googleapis.com/compute/v1/projects/your-project-id/global/firewalls/branch-allow-vpn].
done.Step 2: a test VM in each network
Why: something to ping. Neither VM has a public IP; the only way between them will be the VPN.
# One small test VM in each network, with no public IP: the only way between them will be the VPN
gcloud compute instances create office-vm --zone=us-east1-b --machine-type=e2-micro --subnet=office-sub --no-address --image-family=debian-12 --image-project=debian-cloud > /dev/null 2>&1
gcloud compute instances create branch-vm --zone=us-central1-a --machine-type=e2-micro --subnet=branch-sub --no-address --image-family=debian-12 --image-project=debian-cloud > /dev/null 2>&1
gcloud compute instances list --filter="name~-vm$" --format="table(name,zone.basename(),networkInterfaces[0].networkIP)"NAME ZONE NETWORK_IP
branch-vm us-central1-a 10.20.0.2
office-vm us-east1-b 10.10.0.2Step 3: before the VPN, no route
# Before the VPN: the office VM cannot reach the branch VM (different networks, no route)
until gcloud compute ssh office-vm --zone=us-east1-b --tunnel-through-iap --command="true" > /dev/null 2>&1; do sleep 10; done
gcloud compute ssh office-vm --zone=us-east1-b --tunnel-through-iap --command="ping -c 3 -W 2 10.20.0.2 | tail -2" 2>/dev/null3 packets transmitted, 0 received, 100% packet loss, time 2054ms
100 percent packet loss: two separate networks know nothing about each other.
Step 4: HA VPN gateways and Cloud Routers
Why: one gateway and one Cloud Router per side, each router with its own ASN. Each gateway gets two public IPs (interfaces 0 and 1).
# HA VPN: one gateway per side (each gets TWO public IPs, interface 0 and 1) and one Cloud Router per side for BGP
gcloud compute vpn-gateways create gw-office --network=vpc-office --region=us-east1 > /dev/null
gcloud compute vpn-gateways create gw-branch --network=vpc-branch --region=us-central1 > /dev/null
gcloud compute routers create rt-office --network=vpc-office --region=us-east1 --asn=65001 > /dev/null
gcloud compute routers create rt-branch --network=vpc-branch --region=us-central1 --asn=65002 > /dev/null
gcloud compute vpn-gateways list --format="table(name,region.basename(),vpnInterfaces.len():label=PUBLIC_IPS)"
gcloud compute routers list --filter="name~^rt-" --format="table(name,region.basename(),bgp.asn)"Creating VPN Gateway...
........done.
Creating VPN Gateway...
.........done.
Creating router [rt-office]...
..........done.
Creating router [rt-branch]...
..........done.
NAME REGION PUBLIC_IPS
gw-branch us-central1 2
gw-office us-east1 2
NAME REGION ASN
rt-branch us-central1 65002
rt-office us-east1 65001Step 5: four tunnels
Why: interface 0 to interface 0 and 1 to 1, from both sides. The shared secret is made in the lab with openssl rand and never printed. The peer gateway is given by its full resource path (projects/.../regions/.../vpnGateways/...); with the current gcloud, giving its name plus a separate region flag was refused in our first try.
# Four tunnels (two from each side, one per gateway interface) with one shared secret made in the lab, never printed
openssl rand -base64 24 > .vpn-secret
for i in 0 1; do
gcloud compute vpn-tunnels create office-to-branch-$i --region=us-east1 --vpn-gateway=gw-office --interface=$i --peer-gcp-gateway=projects/$PROJECT_ID/regions/us-central1/vpnGateways/gw-branch --router=rt-office --ike-version=2 --shared-secret="$(cat .vpn-secret)" > /dev/null
gcloud compute vpn-tunnels create branch-to-office-$i --region=us-central1 --vpn-gateway=gw-branch --interface=$i --peer-gcp-gateway=projects/$PROJECT_ID/regions/us-east1/vpnGateways/gw-office --router=rt-branch --ike-version=2 --shared-secret="$(cat .vpn-secret)" > /dev/null
done
gcloud compute vpn-tunnels list --format="table(name,region.basename(),status)"Creating VPN tunnel...
..........done.
Creating VPN tunnel...
..........done.
Creating VPN tunnel...
..........done.
Creating VPN tunnel...
..........done.
NAME REGION STATUS
branch-to-office-0 us-central1 ESTABLISHED
branch-to-office-1 us-central1 FIRST_HANDSHAKE
office-to-branch-0 us-east1 ESTABLISHED
office-to-branch-1 us-east1 FIRST_HANDSHAKETunnels show ESTABLISHED as soon as both sides of a pair exist and agree on the secret; FIRST_HANDSHAKE means the other side was still being created.
Step 6: BGP sessions
Why: each tunnel gets a router interface with a link address, and each router gets a BGP peer on the other side. Then the routers exchange routes by themselves. We wait until both sessions are UP:
# BGP: each tunnel gets a router interface with a link address (169.254.x.x/30) and a BGP peer on the other side
for i in 0 1; do
gcloud compute routers add-interface rt-office --region=us-east1 --interface-name=if-$i --vpn-tunnel=office-to-branch-$i --ip-address=169.254.$i.1 --mask-length=30 > /dev/null
gcloud compute routers add-bgp-peer rt-office --region=us-east1 --peer-name=peer-$i --interface=if-$i --peer-ip-address=169.254.$i.2 --peer-asn=65002 > /dev/null
gcloud compute routers add-interface rt-branch --region=us-central1 --interface-name=if-$i --vpn-tunnel=branch-to-office-$i --ip-address=169.254.$i.2 --mask-length=30 > /dev/null
gcloud compute routers add-bgp-peer rt-branch --region=us-central1 --peer-name=peer-$i --interface=if-$i --peer-ip-address=169.254.$i.1 --peer-asn=65001 > /dev/null
done
for t in $(seq 1 30); do U=$(gcloud compute routers get-status rt-office --region=us-east1 --format="value(result.bgpPeerStatus[].status)" | tr ';' '\n' | grep -c '^UP$'); [ "$U" = 2 ] && break; sleep 10; done
gcloud compute vpn-tunnels list --format="table(name,status,detailedStatus)"
gcloud compute routers get-status rt-office --region=us-east1 --format="table(result.bgpPeerStatus[].name,result.bgpPeerStatus[].status,result.bgpPeerStatus[].numLearnedRoutes)"Updated [https://www.googleapis.com/compute/v1/projects/your-project-id/regions/us-east1/routers/rt-office].
Creating peer [peer-0] in router [rt-office]...
..........done.
Updated [https://www.googleapis.com/compute/v1/projects/your-project-id/regions/us-central1/routers/rt-branch].
Creating peer [peer-0] in router [rt-branch]...
..........done.
Updated [https://www.googleapis.com/compute/v1/projects/your-project-id/regions/us-east1/routers/rt-office].
Creating peer [peer-1] in router [rt-office]...
..........done.
Updated [https://www.googleapis.com/compute/v1/projects/your-project-id/regions/us-central1/routers/rt-branch].
Creating peer [peer-1] in router [rt-branch]...
..........done.
NAME STATUS DETAILED_STATUS
branch-to-office-0 ESTABLISHED Tunnel is up and running.
branch-to-office-1 ESTABLISHED Tunnel is up and running.
office-to-branch-0 ESTABLISHED Tunnel is up and running.
office-to-branch-1 ESTABLISHED Tunnel is up and running.
NAME STATUS NUM_LEARNED_ROUTES
['peer-0', 'peer-1'] ['UP', 'UP'] [1, 1]All four tunnels Tunnel is up and running, both BGP sessions UP, one route learned on each.
Step 7: ping through the VPN
# Through the VPN: the office VM now reaches the branch VM on its private address, and learned the route by BGP
gcloud compute routers get-status rt-office --region=us-east1 --format="value(result.bestRoutes[].destRange)" | tr ';' '\n' | sort -u | sed 's/^/route learned by BGP: /'
gcloud compute ssh office-vm --zone=us-east1-b --tunnel-through-iap --command="ping -c 4 10.20.0.2 | tail -2" 2>/dev/nullroute learned by BGP: 10.20.0.0/24
route learned by BGP: 10.20.0.0/24
4 packets transmitted, 4 received, 0% packet loss, time 3002ms
rtt min/avg/max/mdev = 34.691/35.518/37.712/1.271 msThe office router learned 10.20.0.0/24 (the branch subnet) from BGP over both tunnels, and the office VM reaches the branch VM's private address in about 35 ms (us-east1 to us-central1).

Break it: a tunnel goes down
We delete one tunnel, as if one internet link or one gateway interface failed:
# Break it: one tunnel goes down (we delete it, like a failed internet link). HA VPN keeps traffic on the other one
gcloud compute vpn-tunnels delete office-to-branch-0 --region=us-east1 --quiet 2>&1 | tail -1; sleep 30
gcloud compute routers get-status rt-office --region=us-east1 --format="table(result.bgpPeerStatus[].name,result.bgpPeerStatus[].status)"
gcloud compute ssh office-vm --zone=us-east1-b --tunnel-through-iap --command="ping -c 4 10.20.0.2 | tail -2" 2>/dev/null..........done.
NAME STATUS
['peer-0', 'peer-1'] ['DOWN', 'UP']
4 packets transmitted, 4 received, 0% packet loss, time 3004ms
rtt min/avg/max/mdev = 34.826/35.672/37.821/1.244 ms📝⭐ Recommended by CareerByteCode · Certification examCIPS Level 5 Advanced Diploma in Procurement and Supply PracticeTake the examTake the exam →One BGP session is DOWN, but the ping still gets 4 of 4 replies: traffic moved to the remaining tunnel. That is the "HA" in HA VPN, and it only works because we built two tunnels.
Break it again: the wrong shared secret
We recreate the tunnel with a different secret than the other side uses, a classic mistake when two teams configure the two ends:
# Break it again: recreate the tunnel with a WRONG shared secret. IKE cannot agree, the tunnel never comes up
gcloud compute vpn-tunnels create office-to-branch-0 --region=us-east1 --vpn-gateway=gw-office --interface=0 --peer-gcp-gateway=projects/$PROJECT_ID/regions/us-central1/vpnGateways/gw-branch --router=rt-office --ike-version=2 --shared-secret="wrong-$(openssl rand -hex 6)" > /dev/null
sleep 60; gcloud compute vpn-tunnels describe office-to-branch-0 --region=us-east1 --format="value(status,detailedStatus)"Creating VPN tunnel...
..........done.
FIRST_HANDSHAKE Allocating resources. VPN tunnel will start soon.After 60 seconds the tunnel is still in FIRST_HANDSHAKE: IKE cannot finish because the two sides do not share the same key. (We did not wait longer; a tunnel with a wrong secret does not come up until the secret is fixed.)
Fix it
Recreate the tunnel with the right secret and wait for its BGP session:
# Fix it: recreate the tunnel with the right secret; the BGP peer comes back and both paths are up again
gcloud compute vpn-tunnels delete office-to-branch-0 --region=us-east1 --quiet > /dev/null 2>&1
gcloud compute vpn-tunnels create office-to-branch-0 --region=us-east1 --vpn-gateway=gw-office --interface=0 --peer-gcp-gateway=projects/$PROJECT_ID/regions/us-central1/vpnGateways/gw-branch --router=rt-office --ike-version=2 --shared-secret="$(cat .vpn-secret)" > /dev/null
for t in $(seq 1 30); do U=$(gcloud compute routers get-status rt-office --region=us-east1 --format="value(result.bgpPeerStatus[].status)" | tr ';' '\n' | grep -c '^UP$'); [ "$U" = 2 ] && break; sleep 10; done
gcloud compute routers get-status rt-office --region=us-east1 --format="table(result.bgpPeerStatus[].name,result.bgpPeerStatus[].status)"Creating VPN tunnel...
..........done.
NAME STATUS
['peer-0', 'peer-1'] ['UP', 'UP']Both BGP sessions are UP again.

POC: prove it works
- No route before the VPN, 4 of 4 pings after it, over private addresses (Steps 3 and 7).
- BGP learned the branch range automatically (Step 7).
- With one tunnel gone, traffic kept flowing on the other (Break it).
- A wrong secret kept the tunnel from coming up; the right one fixed it (Break it again, Fix it).
Clean up (delete everything)
Why: tunnels and gateways cost by the hour. We delete in order: tunnels, routers, gateways, VMs, firewall rules, subnets, networks.
# Clean up fast (tunnels cost by the hour): tunnels, routers, gateways, VMs, firewall rules, subnets, networks
for i in 0 1; do gcloud compute vpn-tunnels delete office-to-branch-$i --region=us-east1 --quiet > /dev/null 2>&1; gcloud compute vpn-tunnels delete branch-to-office-$i --region=us-central1 --quiet > /dev/null 2>&1; done
gcloud compute routers delete rt-office --region=us-east1 --quiet > /dev/null 2>&1; gcloud compute routers delete rt-branch --region=us-central1 --quiet > /dev/null 2>&1
gcloud compute vpn-gateways delete gw-office --region=us-east1 --quiet > /dev/null 2>&1; gcloud compute vpn-gateways delete gw-branch --region=us-central1 --quiet > /dev/null 2>&1
gcloud compute instances delete office-vm --zone=us-east1-b --quiet > /dev/null 2>&1; gcloud compute instances delete branch-vm --zone=us-central1-a --quiet > /dev/null 2>&1
for n in office branch; do gcloud compute firewall-rules delete $n-allow-vpn --quiet > /dev/null 2>&1; done
gcloud compute networks subnets delete office-sub --region=us-east1 --quiet > /dev/null 2>&1; gcloud compute networks subnets delete branch-sub --region=us-central1 --quiet > /dev/null 2>&1
gcloud compute networks delete vpc-office vpc-branch --quiet > /dev/null 2>&1; rm -f .vpn-secret
echo "left: $(gcloud compute vpn-tunnels list --format='value(name)' | wc -l) tunnels, $(gcloud compute vpn-gateways list --format='value(name)' | wc -l) gateways, $(gcloud compute routers list --filter='name~^rt-' --format='value(name)' | wc -l) routers, $(gcloud compute instances list --filter='name~-vm$' --format='value(name)' | wc -l) VMs, $(gcloud compute networks list --filter='name~^vpc-' --format='value(name)' | wc -l) networks"WARNING: The following filter keys were not present in any resource : name
left: 0 tunnels, 0 gateways, 0 routers, 0 VMs, 0 networks
FAQ
Classic VPN or HA VPN? HA VPN: two interfaces, a 99.99 percent SLA when set up with two tunnels, and dynamic routing with BGP. Classic VPN is the older option; Google recommends HA VPN for new connections.
How do I connect an office router instead of a second VPC? Create an external VPN gateway object with the office router's public IP, and use --peer-external-gateway in the tunnels. The office side needs IKEv2, the same secret and BGP. We did not test a physical router here.
Why BGP and not static routes? BGP adds and removes routes by itself when a tunnel or a subnet changes, which is what made the failover in Break it automatic.
Can the two networks use the same IP range? No. Overlapping ranges cannot be routed between each other. Plan ranges before you connect sites.
What about Cloud Interconnect? A private physical connection with more bandwidth and no internet in between. VPN is the quick, cheap start; Interconnect is for big, steady traffic.
Troubleshooting
| You see | Why | Fix |
|---|---|---|
Exactly one of (--peer-address ... --peer-gcp-gateway-region) must be specified | Current gcloud refuses name plus region for the peer gateway | Give the full path: projects/PROJECT/regions/REGION/vpnGateways/NAME |
Tunnel stays in FIRST_HANDSHAKE | The other side is missing, or the shared secrets differ | Create both sides; use the same secret |
Tunnel up, BGP DOWN | Link addresses or peer ASN do not match | Each side's peer IP is the other side's interface IP; ASNs must match the routers |
| BGP up, ping fails | Firewall rules do not allow the other site's range | Allow the remote subnet (and ICMP) in each network |
| Ping works one way only | Routes or firewall rules on one side only | Check learned routes on both routers |
What next
- Learn the network basics behind this: Linux Networking Commands Every Cloud Engineer Needs.
- Run a database on a private network: Cloud SQL for PostgreSQL: Create, Connect and Back Up.
- More hands-on posts: ByteBlog. Practice in ByteLabs and see the courses.
Prepared by the CareerByteCode Team from our tested lab, with AI help for the write-up.
⭐ Keep learning with CareerByteCode
Hand-picked to go deeper on this topicGet CareerByteCode Team's next blog by email
One email a week when they publish. Add the weekly Best of ByteBlog too. No spam, unsubscribe with one click.
We email you once to confirm. By subscribing you agree to the privacy policy.
Related blogs
CareerByteCode TeamCilium CNI: eBPF Networking and Security for Kubernetes
What you build: Cilium network policies on GKE Dataplane V2: default deny, dropped packets seen in the kernel, egress control and a DNS mistake fixed. Every step tested.
#advanced
CareerByteCode TeamLinkerd vs Istio: Lightweight Service Mesh Hands-On
What you build: Istio and Linkerd on one GKE cluster, the same app and load measured side by side, and a plain-text caller test. Every step tested.
#beginners
CareerByteCode TeamCloud DNS: Host Your Domain on Google Cloud
What you build: Cloud DNS public and private zones, a caching resolver that serves an old answer until its TTL runs out, and the lower-TTL-first fix. Every step tested.
#beginners
CareerByteCode TeamGoogle Cloud VPC for Beginners: Subnets, Routes and Firewalls
What you build: a custom VPC with two subnets and firewall rules, in your own Google Cloud account. Every step, command and screenshot explained for beginners.
💬 Comments (0)
Top · NewestSign in to comment